Lab 05 · Local fictional budget

Build a context window

Allocate limited capacity across instructions, conversation, a retrieved document, and the answer.

Prepared total capacity300 units are always reserved for the answer.

Requested blocks, in retention orderSystem instructionsRole, constraints, and expected format.180 unitskeptRecent conversationThe most useful recent turns.300 unitskeptRetrieved documentAn excerpt supplied by a retrieval system.520 unitskeptOlder conversationEarlier turns, less closely related to the current request.360 unitsremoved: no space

Input 1000Answer 300

Prepared plan

1 block removed

Input budget: 1100 units. Requested: 1360 units. Theoretical overflow: 260 units.

Limits of this representation

The units are not real token counts.
This strategy keeps blocks in a fixed order; real applications may truncate, summarize, or retrieve differently.
Fitting in the window does not guarantee that the model will use every piece of information equally well.